ACSIS Project

Organization
The ACSIS project is a collaboration of several observatories or laboratories. The project is lead by Dominion Radio Astrophysical Observatory (DRAO) in Canada and done in collaboration with the UK Astronomical Technology Centre (ATC) and the Joint Astronomy Centre (JAC) in Hawaii. The current areas of responsibilities are:

Specifications

• Support 16 channel array receivers.
• A velocity coverage to support Galaxy observations; velocity coverage of minimum 700 km/s (800MHz at 345GHz and 1600MHz at 690GHz)
• Spectra dump time 100ms goal 50ms. The fast dump rate allows ACSIS to use the full speed of the chopping secondary and to do rapid large-scale mapping.
• Real time data reduction giving the observer control of the operation and enabling reduced data to be exported.
• Fast sampling to avoid extensive IF circuits and opportunities for platforming.
• Sampler driven at constant power levels to avoid platforming.
• Total power detectors linear enough to restore the total power information and to be used for continuum observations.

Design
At the core of ACSIS are 32 Down Converter Modules (DCMs), samplers and correlator boards. A DCM selects a 1 GHz or 250 MHz wide band in the IF frequency rage of 3.2 to 7.5 GHz and down converts the band to base band. After a 2 bit 3 state sampling with 2 GHz the digital signal is feed to a correlator board. With a 2 GHz sampling rate a 1 GHz band can be reconstructed. However, due to filtering requirements the effective bandwidth of each system is less than 1 GHz but at least 800 MHz. In the case of a 250MHz wide IF band the effective usable bandwidth is at least 200 MHz. In the following these bands are denoted 1 GHz or 250 MHz but it should be understood that the effective bandwidth is less.

The optics of the JCMT is not designed to allow the use of several instruments at once. Hence, the maximum number of IF channels are 16 after the arrival of HARP-B. This is the reason the ACSIS IF system at the JCMT only will have 16 feeds. An IF switch on the Nasmyth platform switches between the HARP-B array and the cabin receivers. The switch is followed by IF cables down to ACSIS on the carousel floor. The cables feed two ACSIS IF-racks each containing 2 quadrant switches, 16 DCMs and 16 samplers. A quadrant switch receives 4 IF cables and feeds 8 DCMs. If 16 pixels are used each IF input is feed to two DCMs. Normally these DCMs are configured to observe two adjacent IF frequency ranges creating a contiguous coverage of up to effectively 1.6 GHz. However, it is also possible to have the bands split and operating with different bandwidths (1GHz/250MHz). Each feed will then have several subsystems. By reducing the number of feeds or pixels on the sky more DCMs can be used for a single IF feed. These modes will be used with the cabin receivers or a subset of HARP-Bs array.

Technical details of interest - the DCM has total power detectors used for automatic gain control. This keeps the input levels to the sampler's constant eliminating platforming. Other total power detectors in the DCMs are read out synchronized with the data acquisition to restore the amplitude information. The total power data will also be used for continuum observations and pointing and focusing. The correlator boards operate at 62.5 MHz. The 2 GHz sample rate is supported by recording the data and play it back at a reduced rate to 32 sub correlators. In 250 MHz mode 4 times more channels becomes available since only 8 sub correlators are needed per DCM. The same sampler can feed two correlator boards increasing the resolution to the prize of only using 16 DCMs/samplers. Since only 16 feeds is supported at the JCMT this only reduces the bandwidth in this mode not the number of available feeds. There is only 4 independent LOs used to select the 1GHz or 250 MHz IF band excluding an 8x1GHz and similar mode.

Summary of useful configurations at the JCMT (max 16 IF feeds)

Data acquisition and reduction
The target for the data reduction is to be able to reduce 20 spectra per second. This will allow observations to be synchronized with the chopper at high speed for improved baseline and total power stability. ACSIS will replace the present backend for continuum observations so high stability is needed pointing, focus and other continuum observations. During mapping observations the data is gridded on a data "cube" in real time for control of the observations. The cube display is part of the real time display subsystem.

The aim for the data reduction system is to generate data that the observer will use for further reduction or publication. It is anticipated that the observer will leave with the reduced data. The raw data amount can otherwise be overwhelming and few people would have the ability to re-reduce the data away from the JCMT. There should therefore be no reason to rereduce the data later. Poor quality data needs to be detected and actions taken - a mixture of reobserving and rereduction. The real time display system aids the operator/observer in gauge the data quality. The ACSIS real time display has three basic components. In addition to the cube display there is a chart recorder display that displays up to 16 different parameters and a spectrum display for non-gridded spectra.

Technical details
8 correlator computers read out the 32-correlator boards. These computers add up the data from the sub correlators, remove bad channels and truncate the data. The data is then passed to the data reduction computers were the data is synched with information messages from the receiver, IF system secondary, antenna etc. The data messages from the different sources are tagged with sequence numbers to facilitate the synchronization. This process allows the data acquisition to continue while the data from the previous observations is processed. When all messages has been received the data is passed the reducer tasks. The reducer task generates calibrated spectra out of the individual observations. The finished spectra are then passed to the girder task, which is responsible for gridding, the data onto a map cube. The data reduction is likely to be run on 8-16 dual processor high performance Linux PCs. While the reducer task CPU receives data from one or two feed per CPU the gridder task will need to combine data from all the feeds. However, the computing load still has to be shared over the cluster. This is achieved by dividing the regridding in frequency slices, which are divided between the CPUs. Substantial amounts of inter CPU communication is needed at this point.